Predicting Power Outage at Low-Lying Area Substations During Storm Surge Disasters Using Multi-Grained Cascaded Forest

Previous studies and historical events show that storm surge disasters can lead to power outages. This calls for advanced situational awareness technologies for ensuring electrical grid security during storm surge events. However, existing efforts are designed for typhoon disasters and fail to consider the impact of storm surges on substations. To address this issue, we propose the first grid security situation-aware early warning framework to address the power outage problem at substations during storm surge disasters. Our approach leverages geospatial meteorological data and information about power equipment to accurately predict the severity of storm surges and issue timely warnings regarding potential power outages at substations. We employ the ArcGIS 10.8 geographic information system desktop platform to collect high-resolution LiDAR data on coastal cities, enabling the creation of digital twins that offer comprehensive situational awareness. Furthermore, by inputting geographic information data, meteorological data and power equipment information to a multi-granularity ensemble forest model, we achieve deterministic and probabilistic power outage warnings for substations during storm surges. This enables a comprehensive understanding and visualization of the situation. We validate our methodology by applying it to analyze historical typhoon data from a coastal city in China. The results demonstrate the effectiveness and reliable accuracy across datasets of varying scales of our approach, which overcomes issues related to overfitting and avoids frequent network structure adjustments or parameter tuning. Our predicted deterministic and uncertain risk-free rates have an error of less than 0.5%.